yanquis1972 wrote:don, can't/haven't you experimented with a 'komodo tactical' to, at very least, figure out what parts of the code are causing the poor results? as i said, from my perspective, this seems like a very good thing, as it suggests a lot of elo points are to be gained if tactics can be improved w/out sacrificing 'intelligence', or at least, too much of it. while i know being the #1 engine is your goal, part of komodo's appeal to me has always been its knowledge, or what comes across as such.
it seems to me like, almost all of the time, the top engine in the world is either the best, or very close to the best, tactically. again, rybka was a huge exception, but vas found a way to not only keep but add knowledge, & -massively- improve rybka tactically at the same time.
perhaps with faster hardware (the above test was run on a 12 core monster), this is becoming less relevant, but i also recall people saying the same some 15 years ago (all CS TAL II needs to be competitve is an ultrafast machine, etc), so i tend to disregard this angle.
We have no desire for Komodo to have a reputation for being weak in tactics, so we have tried to improve this - but we don't obsess about it because it's abundantly clear that solving problem sets does not correlate very well with actual strength of play.
We have tried more aggressive tactical extensions, less pruning, more pruning and other things. The easiest way to get a big tactical boost is to randomize the root move ordering. Does that seem like a good thing to you? Actually, it had only a very minor impact on the ELO, but it was clearly measurable. But it seems the path to getting really good scores on tactical sets is to be willing to give up a few ELO. We could get fantastic scores if we were willing to give up just 5 or 10 ELO, but we are not willing to do that.
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
, even if not by much.